mirror of
https://github.com/gentoo-mirror/gentoo.git
synced 2026-09-24 04:59:14 -07:00
sci-ml/caffe2: drop 2.4.1-r7
Signed-off-by: Alfredo Tupone <tupone@gentoo.org>
This commit is contained in:
Alfredo Tupone
@@ -1,8 +1,6 @@
|
||||
DIST caffe2-patches-20240809.tar.gz 15242 BLAKE2B 77503c61487e7d85cca5afcab9a6e638f9833a70861845638cf1b62bc492d7b6650e6db81d53ebb2f39c6313509250d339f725f04d03ec6dd23dd0cf70843d8c SHA512 74b3b0b6671b655ecac93f7436c4ed7cb0157a83aafbf6afcc0811e11cef341cd8f638db1a111bcbb01e1a6dd4daf3a36b96d7a8ce90f04c2fa091bd6e3a142b
|
||||
DIST composable_kernel-50ee4267.tar.gz 4194795 BLAKE2B b3c97d98a0c9e4620fdae3d30006edf55cc60ffa7f8518f6acb8d808647bc4de362c2e2b7e974686503fa2c7f359b6981cfbda74e40cc1bad4d351c5d2ff92e1 SHA512 9fc6f5f15556f020414b4567520329ef762209a82411a246c2bc1240a9fed2669f7fcb982cf773e3e9561bf9a2c557dba82b8b469d2e5844e679e2f5ab7c3e17
|
||||
DIST composable_kernel-8086bbe3.tar.gz 4418862 BLAKE2B b710e3d4586899443ec01044dad19fd2f992c351e2f65ba526dfcc47cc65c095beaf8ac21a8f71c02a0eb524d364e817b27241a9198884f2bdae9924b51e24e4 SHA512 8410b5a1c864d71f3034ef0d9d1245078856d09cc191faec59856c229bf11d89ae291036d735cb5cec4f1d72e6e9e8f6921833147f9619d30cfab8722d3a9f63
|
||||
DIST flash-attention-2.7.4.gh.tar.gz 5841323 BLAKE2B 432999d763f2b3d732580ddfea5d3e01370351db0656546259a5e500a07516dd03c98828bfb55855dabe4adc651033b5d97ea4725ca46158b9970f0fbc662710 SHA512 05a4afb09e666f7404d6a3f8b5256e7bed6eba60a6f1bde2b7dbb96d318975f0b458c2521c7a38d88e97b6e4c27f29077cf787849daf82586e33f43a3d9a84b3
|
||||
DIST pytorch-2.4.1.tar.gz 115029469 BLAKE2B c2909ff27d527bc57cba56b780d3b8cd07a043ab045caa6c6b27857a16f9ad10aaab2116b26226b1e46ee08ffb44007965d914464418e4ae14ca48c3f3f383bb SHA512 7e9b4485e242eaf0d648765c6621d73d95e7107b766646a098175436d1ab2e2b864badd0757a3bab6b7c318233f2120bad9ac07b39bb9e357897919580c87631
|
||||
DIST pytorch-2.5.1.tar.gz 116091366 BLAKE2B 7838b17562b94ffc7d798031348689db607dd5eae2a3c35be365972e2b52a2c1b12067068d5aca5ab00cf0977d9c2c3c9ae5337d69534c864c732e6256cbeef6 SHA512 a913a466324a65fa3d79c5e9ad4d605fc7976f0134fda2f81aaa3cea29d56926604999b8a238759646d211e63b47bbb446cdffa86ca8defd8159f11e30301289
|
||||
DIST pytorch-2.6.0.tar.gz 119594438 BLAKE2B 3152eb341cf42295e147e59625beb9c06608aa4b78f9618c1c0024b10c1c767715d07fe8c4be52d029ac47f808cd0d5e65c9530ec90d951a64b993083b4067ad SHA512 a70da80ff09d226085e18228132cf6bb236ad8cc47eed52375d0d2a615f09dd33849da947270b5670c184eab60cb8e2adf11d801babfbda7aa621400501d07b0
|
||||
DIST pytorch-2.7.0.tar.gz 50197290 BLAKE2B 2a317d1e9b0d8876f1593382246cd9f786eff3c1b8602353c5e0010dc8414720c5de61886361843a0c33268830c784963a89b410b361e1b67636e652f6a6a2eb SHA512 63eb0363ea68d23567f5524ee8b51756d9302bbe1cbefa367335ab5ebe652523dba75fa417ea3e7eedfc67aa4bef1434c8b7e3dfde2152061b91b6e489763a55
|
||||
|
||||
@@ -1,305 +0,0 @@
|
||||
# Copyright 2022-2025 Gentoo Authors
|
||||
# Distributed under the terms of the GNU General Public License v2
|
||||
|
||||
EAPI=8
|
||||
|
||||
PYTHON_COMPAT=( python3_{10..13} )
|
||||
ROCM_VERSION=6.1
|
||||
inherit python-single-r1 cmake cuda flag-o-matic prefix rocm toolchain-funcs
|
||||
|
||||
MYPN=pytorch
|
||||
MYP=${MYPN}-${PV}
|
||||
|
||||
DESCRIPTION="A deep learning framework"
|
||||
HOMEPAGE="https://pytorch.org/"
|
||||
SRC_URI="https://github.com/pytorch/${MYPN}/archive/refs/tags/v${PV}.tar.gz
|
||||
-> ${MYP}.tar.gz
|
||||
https://dev.gentoo.org/~tupone/distfiles/${PN}-patches-20240809.tar.gz"
|
||||
|
||||
S="${WORKDIR}"/${MYP}
|
||||
|
||||
LICENSE="BSD"
|
||||
SLOT="0"
|
||||
KEYWORDS="~amd64"
|
||||
IUSE="cuda distributed fbgemm flash gloo mkl mpi nnpack +numpy onednn openblas opencl openmp qnnpack rocm xnnpack"
|
||||
RESTRICT="test"
|
||||
REQUIRED_USE="
|
||||
${PYTHON_REQUIRED_USE}
|
||||
mpi? ( distributed )
|
||||
gloo? ( distributed )
|
||||
?? ( cuda rocm )
|
||||
rocm? (
|
||||
|| ( ${ROCM_REQUIRED_USE} )
|
||||
!flash
|
||||
)
|
||||
"
|
||||
|
||||
# CUDA 12 not supported yet: https://github.com/pytorch/pytorch/issues/91122
|
||||
RDEPEND="
|
||||
${PYTHON_DEPS}
|
||||
dev-cpp/abseil-cpp:=
|
||||
dev-cpp/gflags:=
|
||||
>=dev-cpp/glog-0.5.0
|
||||
dev-libs/cpuinfo
|
||||
dev-libs/libfmt:=
|
||||
dev-cpp/opentelemetry-cpp
|
||||
dev-libs/protobuf:=
|
||||
dev-libs/pthreadpool
|
||||
dev-libs/sleef[cpu_flags_x86_avx512f(+),cpu_flags_x86_avx(+)]
|
||||
dev-libs/sleef[cpu_flags_x86_sse3(+),cpu_flags_x86_ssse3(+)]
|
||||
dev-libs/sleef[cpu_flags_x86_sse4_1(+),cpu_flags_x86_sse4_2(+)]
|
||||
virtual/lapack
|
||||
sci-ml/onnx
|
||||
sci-ml/foxi
|
||||
cuda? (
|
||||
dev-libs/cudnn
|
||||
>=sci-ml/cudnn-frontend-1.0.3:0/8
|
||||
<dev-util/nvidia-cuda-toolkit-12.5:=[profiler]
|
||||
)
|
||||
fbgemm? ( sci-ml/FBGEMM )
|
||||
gloo? ( sci-ml/gloo[cuda?] )
|
||||
mpi? ( virtual/mpi )
|
||||
nnpack? ( sci-ml/NNPACK )
|
||||
numpy? ( $(python_gen_cond_dep '
|
||||
dev-python/numpy[${PYTHON_USEDEP}]
|
||||
') )
|
||||
onednn? ( sci-ml/oneDNN )
|
||||
opencl? ( virtual/opencl )
|
||||
qnnpack? (
|
||||
!sci-libs/QNNPACK
|
||||
sci-ml/gemmlowp
|
||||
)
|
||||
rocm? (
|
||||
=dev-util/hip-6.1*
|
||||
=dev-libs/rccl-6.1*
|
||||
=sci-libs/rocThrust-6.1*
|
||||
=sci-libs/rocPRIM-6.1*
|
||||
=sci-libs/hipBLAS-6.1*
|
||||
=sci-libs/hipFFT-6.1*
|
||||
=sci-libs/hipSPARSE-6.1*
|
||||
=sci-libs/hipRAND-6.1*
|
||||
=sci-libs/hipCUB-6.1*
|
||||
=sci-libs/hipSOLVER-6.1*
|
||||
=sci-libs/miopen-6.1*
|
||||
=dev-util/roctracer-6.1*
|
||||
|
||||
=sci-libs/hipBLASLt-6.1*
|
||||
amdgpu_targets_gfx90a? ( =sci-libs/hipBLASLt-6.1*[amdgpu_targets_gfx90a] )
|
||||
amdgpu_targets_gfx940? ( =sci-libs/hipBLASLt-6.1*[amdgpu_targets_gfx940] )
|
||||
amdgpu_targets_gfx941? ( =sci-libs/hipBLASLt-6.1*[amdgpu_targets_gfx941] )
|
||||
amdgpu_targets_gfx942? ( =sci-libs/hipBLASLt-6.1*[amdgpu_targets_gfx942] )
|
||||
)
|
||||
distributed? (
|
||||
sci-ml/tensorpipe[cuda?]
|
||||
dev-cpp/cpp-httplib
|
||||
)
|
||||
xnnpack? ( sci-ml/XNNPACK )
|
||||
mkl? ( sci-libs/mkl )
|
||||
openblas? ( sci-libs/openblas )
|
||||
"
|
||||
DEPEND="
|
||||
${RDEPEND}
|
||||
dev-libs/clog
|
||||
dev-libs/psimd
|
||||
dev-libs/FXdiv
|
||||
dev-libs/pocketfft
|
||||
dev-libs/flatbuffers
|
||||
sci-ml/FP16
|
||||
sci-ml/kineto
|
||||
$(python_gen_cond_dep '
|
||||
dev-python/pybind11[${PYTHON_USEDEP}]
|
||||
dev-python/pyyaml[${PYTHON_USEDEP}]
|
||||
dev-python/typing-extensions[${PYTHON_USEDEP}]
|
||||
')
|
||||
cuda? ( <=dev-libs/cutlass-3.4.1 )
|
||||
onednn? ( sci-ml/ideep )
|
||||
"
|
||||
|
||||
PATCHES=(
|
||||
../patches/${PN}-2.4.0-gentoo.patch
|
||||
../patches/${PN}-2.4.0-install-dirs.patch
|
||||
../patches/${PN}-1.12.0-glog-0.6.0.patch
|
||||
../patches/${PN}-1.13.1-tensorpipe.patch
|
||||
../patches/${PN}-2.3.0-cudnn_include_fix.patch
|
||||
../patches/${PN}-2.1.2-fix-rpath.patch
|
||||
../patches/${PN}-2.4.0-fix-openmp-link.patch
|
||||
../patches/${PN}-2.4.0-rocm-fix-std-cpp17.patch
|
||||
../patches/${PN}-2.2.2-musl.patch
|
||||
../patches/${PN}-2.4.0-exclude-aotriton.patch
|
||||
../patches/${PN}-2.3.0-fix-rocm-gcc14-clamp.patch
|
||||
../patches/${PN}-2.3.0-fix-libcpp.patch
|
||||
"${FILESDIR}"/${PN}-2.4.0-libfmt-11.patch
|
||||
"${FILESDIR}"/${PN}-2.4.0-cpp-httplib.patch
|
||||
"${FILESDIR}"/${PN}-2.4.0-cstdint.patch
|
||||
)
|
||||
|
||||
src_prepare() {
|
||||
filter-lto #bug 862672
|
||||
sed -i \
|
||||
-e "/third_party\/gloo/d" \
|
||||
cmake/Dependencies.cmake \
|
||||
|| die
|
||||
cmake_src_prepare
|
||||
pushd torch/csrc/jit/serialization || die
|
||||
flatc --cpp --gen-mutable --scoped-enums mobile_bytecode.fbs || die
|
||||
popd
|
||||
# prefixify the hardcoded paths, after all patches are applied
|
||||
hprefixify \
|
||||
aten/CMakeLists.txt \
|
||||
caffe2/CMakeLists.txt \
|
||||
cmake/Metal.cmake \
|
||||
cmake/Modules/*.cmake \
|
||||
cmake/Modules_CUDA_fix/FindCUDNN.cmake \
|
||||
cmake/Modules_CUDA_fix/upstream/FindCUDA/make2cmake.cmake \
|
||||
cmake/Modules_CUDA_fix/upstream/FindPackageHandleStandardArgs.cmake \
|
||||
cmake/public/LoadHIP.cmake \
|
||||
cmake/public/cuda.cmake \
|
||||
cmake/Dependencies.cmake \
|
||||
torch/CMakeLists.txt \
|
||||
CMakeLists.txt
|
||||
|
||||
if use rocm; then
|
||||
sed -e "s:/opt/rocm:/usr:" \
|
||||
-e "s:lib/cmake:$(get_libdir)/cmake:g" \
|
||||
-e "s/HIP 1.0/HIP 1.0 REQUIRED/" \
|
||||
-i cmake/public/LoadHIP.cmake || die
|
||||
|
||||
ebegin "HIPifying cuda sources"
|
||||
${EPYTHON} tools/amd_build/build_amd.py || die
|
||||
eend $?
|
||||
fi
|
||||
}
|
||||
|
||||
src_configure() {
|
||||
if use cuda && [[ -z ${TORCH_CUDA_ARCH_LIST} ]]; then
|
||||
ewarn "WARNING: caffe2 is being built with its default CUDA compute capabilities: 3.5 and 7.0."
|
||||
ewarn "These may not be optimal for your GPU."
|
||||
ewarn ""
|
||||
ewarn "To configure caffe2 with the CUDA compute capability that is optimal for your GPU,"
|
||||
ewarn "set TORCH_CUDA_ARCH_LIST in your make.conf, and re-emerge caffe2."
|
||||
ewarn "For example, to use CUDA capability 7.5 & 3.5, add: TORCH_CUDA_ARCH_LIST=7.5 3.5"
|
||||
ewarn "For a Maxwell model GPU, an example value would be: TORCH_CUDA_ARCH_LIST=Maxwell"
|
||||
ewarn ""
|
||||
ewarn "You can look up your GPU's CUDA compute capability at https://developer.nvidia.com/cuda-gpus"
|
||||
ewarn "or by running /opt/cuda/extras/demo_suite/deviceQuery | grep 'CUDA Capability'"
|
||||
fi
|
||||
|
||||
local mycmakeargs=(
|
||||
-DBUILD_CUSTOM_PROTOBUF=OFF
|
||||
-DBUILD_SHARED_LIBS=ON
|
||||
|
||||
-DUSE_CCACHE=OFF
|
||||
-DUSE_CUDA=$(usex cuda)
|
||||
-DUSE_DISTRIBUTED=$(usex distributed)
|
||||
-DUSE_MPI=$(usex mpi)
|
||||
-DUSE_FAKELOWP=OFF
|
||||
-DUSE_FBGEMM=$(usex fbgemm)
|
||||
-DUSE_FLASH_ATTENTION=$(usex flash)
|
||||
-DUSE_MEM_EFF_ATTENTION=OFF
|
||||
-DUSE_GFLAGS=ON
|
||||
-DUSE_GLOG=ON
|
||||
-DUSE_GLOO=$(usex gloo)
|
||||
-DUSE_KINETO=OFF # TODO
|
||||
-DUSE_MAGMA=OFF # TODO: In GURU as sci-libs/magma
|
||||
-DUSE_MKLDNN=$(usex onednn)
|
||||
-DUSE_NNPACK=$(usex nnpack)
|
||||
-DUSE_XNNPACK=$(usex xnnpack)
|
||||
-DUSE_SYSTEM_XNNPACK=$(usex xnnpack)
|
||||
-DUSE_TENSORPIPE=$(usex distributed)
|
||||
-DUSE_PYTORCH_QNNPACK=$(usex qnnpack)
|
||||
-DUSE_NUMPY=$(usex numpy)
|
||||
-DUSE_OPENCL=$(usex opencl)
|
||||
-DUSE_OPENMP=$(usex openmp)
|
||||
-DUSE_ROCM=$(usex rocm)
|
||||
-DUSE_SYSTEM_CPUINFO=ON
|
||||
-DUSE_SYSTEM_PYBIND11=ON
|
||||
-DUSE_UCC=OFF
|
||||
-DUSE_VALGRIND=OFF
|
||||
-DPython_EXECUTABLE="${PYTHON}"
|
||||
-DUSE_ITT=OFF
|
||||
-DUSE_SYSTEM_PTHREADPOOL=ON
|
||||
-DUSE_SYSTEM_PSIMD=ON
|
||||
-DUSE_SYSTEM_FXDIV=ON
|
||||
-DUSE_SYSTEM_FP16=ON
|
||||
-DUSE_SYSTEM_GLOO=ON
|
||||
-DUSE_SYSTEM_ONNX=ON
|
||||
-DUSE_SYSTEM_SLEEF=ON
|
||||
-DUSE_PYTORCH_METAL=OFF
|
||||
-DUSE_XPU=OFF
|
||||
|
||||
-Wno-dev
|
||||
-DTORCH_INSTALL_LIB_DIR="${EPREFIX}"/usr/$(get_libdir)
|
||||
-DLIBSHM_INSTALL_LIB_SUBDIR="${EPREFIX}"/usr/$(get_libdir)
|
||||
)
|
||||
|
||||
if use mkl; then
|
||||
mycmakeargs+=(-DBLAS=MKL)
|
||||
elif use openblas; then
|
||||
mycmakeargs+=(-DBLAS=OpenBLAS)
|
||||
else
|
||||
mycmakeargs+=(-DBLAS=Generic -DBLAS_LIBRARIES=)
|
||||
fi
|
||||
|
||||
if use cuda; then
|
||||
addpredict "/dev/nvidiactl" # bug 867706
|
||||
addpredict "/dev/char"
|
||||
addpredict "/proc/self/task" # bug 926116
|
||||
|
||||
mycmakeargs+=(
|
||||
-DUSE_CUDNN=ON
|
||||
-DTORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5 7.0}"
|
||||
-DUSE_NCCL=OFF # TODO: NVIDIA Collective Communication Library
|
||||
-DCMAKE_CUDA_FLAGS="$(cuda_gccdir -f | tr -d \")"
|
||||
)
|
||||
elif use rocm; then
|
||||
export PYTORCH_ROCM_ARCH="$(get_amdgpu_flags)"
|
||||
|
||||
mycmakeargs+=(
|
||||
-DUSE_NCCL=ON
|
||||
-DUSE_SYSTEM_NCCL=ON
|
||||
)
|
||||
|
||||
# ROCm libraries produce too much warnings
|
||||
append-cxxflags -Wno-deprecated-declarations -Wno-unused-result
|
||||
|
||||
if tc-is-clang; then
|
||||
# fix mangling in LLVM: https://github.com/llvm/llvm-project/issues/85656
|
||||
append-cxxflags -fclang-abi-compat=17
|
||||
fi
|
||||
fi
|
||||
|
||||
if use onednn; then
|
||||
mycmakeargs+=(
|
||||
-DUSE_MKLDNN=ON
|
||||
-DMKLDNN_FOUND=ON
|
||||
-DMKLDNN_LIBRARIES=dnnl
|
||||
-DMKLDNN_INCLUDE_DIR="${ESYSROOT}/usr/include/oneapi/dnnl"
|
||||
)
|
||||
fi
|
||||
|
||||
cmake_src_configure
|
||||
|
||||
# do not rerun cmake and the build process in src_install
|
||||
sed '/RERUN/,+1d' -i "${BUILD_DIR}"/build.ninja || die
|
||||
}
|
||||
|
||||
python_install() {
|
||||
python_domodule python/caffe2
|
||||
python_domodule python/torch
|
||||
ln -s ../../../../../include/torch \
|
||||
"${D}$(python_get_sitedir)"/torch/include/torch || die # bug 923269
|
||||
}
|
||||
|
||||
src_install() {
|
||||
cmake_src_install
|
||||
|
||||
# Used by pytorch ebuild
|
||||
insinto "/var/lib/${PN}"
|
||||
doins "${BUILD_DIR}"/CMakeCache.txt
|
||||
|
||||
rm -rf python
|
||||
mkdir -p python/torch/include || die
|
||||
mv "${ED}"/usr/lib/python*/site-packages/caffe2 python/ || die
|
||||
cp torch/version.py python/torch/ || die
|
||||
python_install
|
||||
}
|
||||
@@ -1,10 +0,0 @@
|
||||
--- a/caffe2/utils/string_utils.cc 2024-09-05 08:29:06.930438069 +0200
|
||||
+++ b/caffe2/utils/string_utils.cc 2024-09-05 08:29:28.398137596 +0200
|
||||
@@ -3,6 +3,7 @@
|
||||
#include <algorithm>
|
||||
#include <sstream>
|
||||
#include <vector>
|
||||
+#include <cstdint>
|
||||
|
||||
namespace caffe2 {
|
||||
|
||||
@@ -1,44 +0,0 @@
|
||||
Fix build against libfmt-11
|
||||
|
||||
https://github.com/pytorch/pytorch/commit/83eedf66b9e7f52323d9f45c5dfaa64472452595
|
||||
https://github.com/pytorch/pytorch/pull/130628
|
||||
|
||||
From 83eedf66b9e7f52323d9f45c5dfaa64472452595 Mon Sep 17 00:00:00 2001
|
||||
From: Aaron Gokaslan <aaronGokaslan@gmail.com>
|
||||
Date: Tue, 16 Jul 2024 06:12:08 +0000
|
||||
Subject: [PATCH] Update libfmt submodule to 11.0.1 (#130628)
|
||||
|
||||
Update libfmt to 11.0.1 reopen of https://github.com/pytorch/pytorch/pull/129962. Requires a kineto update and moves fmt::join into a separate include so added it where necessary.
|
||||
|
||||
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130628
|
||||
Approved by: https://github.com/aaronenyeshi
|
||||
--- a/torch/csrc/distributed/c10d/socket.cpp
|
||||
+++ b/torch/csrc/distributed/c10d/socket.cpp
|
||||
@@ -32,6 +32,7 @@ C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wdeprecated")
|
||||
#include <fmt/chrono.h>
|
||||
C10_DIAGNOSTIC_POP()
|
||||
#include <fmt/format.h>
|
||||
+#include <fmt/ranges.h>
|
||||
|
||||
#include <torch/csrc/distributed/c10d/error.h>
|
||||
#include <torch/csrc/distributed/c10d/exception.h>
|
||||
--- a/torch/csrc/profiler/standalone/execution_trace_observer.cpp
|
||||
+++ b/torch/csrc/profiler/standalone/execution_trace_observer.cpp
|
||||
@@ -10,6 +10,7 @@
|
||||
#endif // _WIN32
|
||||
|
||||
#include <fmt/format.h>
|
||||
+#include <fmt/ranges.h>
|
||||
#include <chrono>
|
||||
#include <cmath>
|
||||
#include <fstream>
|
||||
--- a/torch/csrc/profiler/util.cpp
|
||||
+++ b/torch/csrc/profiler/util.cpp
|
||||
@@ -5,6 +5,7 @@
|
||||
#include <c10/util/ArrayRef.h>
|
||||
#include <c10/util/irange.h>
|
||||
#include <fmt/format.h>
|
||||
+#include <fmt/ranges.h>
|
||||
|
||||
#ifdef USE_KINETO
|
||||
#include <libkineto.h>
|
||||
Reference in New Issue
Block a user